DistDebug-Bench
Can LLM agents pinpoint the code behind 120 real bugs in 11 distributed systems?
I study how complex systems can detect when they are behaving incorrectly—even when correctness is not explicitly defined.
Lead-authored research
Comfey
Agentic triage for production cloud incidents · co-first author
TrainCheck
Proactive checks for silent errors in machine-learning training.
Xpert
LLM-based query recommendations for production incident management.
Recent
- Comfey accepted to the FSE Industry Track.
- TrainCheck at PyTorch Conference 2025.
Advised by Prof. Ryan Huang.